Nonlinear Dyn (2020) 102:1939–1962
https://doi.org/10.1007/s11071-020-06020-8
ORIGINAL PAPER
On the optimization of n-sub-step composite time
integration methods
Huimin Zhang · Runsen Zhang ·
Yufeng Xing · Pierangelo Masarati
Received: 12 May 2020 / Accepted: 12 October 2020 / Published online: 24 October 2020
© The Author(s) 2020
Abstract A family of n-sub-step composite time inte-
gration methods, which employs the trapezoidal rule
in the first n − 1 sub-steps and a general formula in
the last one, is discussed in this paper. A universal
approach to optimize the parameters is provided for
any cases of n ≥ 2, and two optimal sub-families of
the method are given for different purposes. From lin-
ear analysis, the first sub-family can achieve nth-order
accuracy and unconditional stability with controllable
algorithmic dissipation, so it is recommended for high-
accuracy purposes. The second sub-family has second-
order accuracy, unconditional stability with control-
lable algorithmic dissipation, and it is designed for
heuristic energy-conserving purposes, by preserving as
much low-frequency content as possible. Finally, some
illustrative examples are solved to check the perfor-
mance in linear and nonlinear systems.
Keywords n-Sub-step composite method · Optimiza-
tion · High-accuracy · Energy-conserving
Huimin Zhang · Runsen Zhang · Yufeng Xing
School of Aeronautic Science and Engineering, Beihang
University, Beijing 100083, China
Huimin Zhang (B) · Runsen Zhang · Pierangelo Masarati
Dipartimento di Scienze e Tecnologie Aerospaziali,
Politecnico di Milano, 20156 Milan, Italy
e-mail: huimin.zhang@polimi.it
1 Introduction
Direct time integration methods are frequently used
to predict accurate numerical responses for general
dynamic problems after spatial discretization. Driven
by the pursuit of desirable properties, including higher
accuracy and efficiency, robust stability, and many oth-
ers, a number of excellent methods were proposed in
the past decades.
In terms of the formulations, existing methods are
generally classified into explicit and implicit schemes.
Explicit methods are mostly used in wave propagation
problems, as their conditional stability limits the allow-
able time step size to the highest system frequency.
Implicit methods have fewer restrictions on the prob-
lems to be solved due to the unconditional stability, but
they require more computational efforts per step.
In another way, the integration methods can also
be divided into single-step, multi-sub-step and multi-
step techniques. The single-step methods only adopt
the states of the last step to predict the current one,
while the multi-sub-step methods also need the states
at the intermediate collocation points, and the multi-
step methods require the states of more than one pre-
vious step. Each of them has specific advantages and
disadvantages.
From the literature, representative single-step meth-
ods include the Newmark method [25], the HHT-α
method (by Hilbert, Hughes, and Taylor) [17], the
WBZ-α method (by Wood, Bossak, and Zienkiewicz)
[29], the generalized-α method [9], the GSSSS (gener-
123
1940
alized single-step single-solve) method [34], and many
others [28]. These single-step methods were proved to
be spectrally identical to the linear multi-step methods
[34], so they suffer from the Dahlquist’s barrier [10],
which states that the methods of higher than second-
order accuracy cannot be unconditionally stable. There-
fore, the methods mentioned above are all second-order
accurate and unconditionally stable; some of them can
also provide controllable algorithmic dissipation.
In the multi-step class, the Dahlquist’s barrier cer-
tainly works, but in terms of accuracy, the linear two-
step method [24,33] is superior to most existing single-
step methods under the same degree of algorithmic
dissipation. In this class, BDFs (backward differenti-
ation formulas) [11,16] also represent a widely-used
branch, particularly useful for stiff problems owing
to the strong algorithmic dissipation. These popular
multi-step methods are also second-order accurate and
unconditionally stable. However, the multi-step meth-
ods are not self-starting, so another method has to be
also used to solve the initial steps, which makes the
multi-step methods not as convenient to use as the
single-step ones.
The multi-sub-step methods, also known as multi-
stage methods, allow more possibilities in terms of
properties. The most representative method is the
famous Runge–Kutta family [6,7,19], which can be
designed to be arbitrarily higher-order accurate and
unconditionally stable by choosing proper parame-
ters and enough stages. Besides, Fung [12–15] pro-
vided some methods to reproduce the generalized Padé
approximation. These methods can reach up to 2nth-
order accuracy by employing n sampling grid points
per step, but the dimension of the implicit equation to
be solved is n times that of the original, resulting in
huge computational costs. In the multi-sub-step class,
the composite methods [3], which divide each step into
several sub-steps and employ different methods in each
sub-step, have received a lot of attention in recent years.
Based on Bank et al.’s work [1], Bathe et al. [3]
introduced the concept of the n-sub-step composite
method by utilizing the trapezoidal rule in the first
n − 1 sub-steps and the (n + 1)-point backward dif-
ference scheme at the end of the step. The two-sub-
step scheme is known as the Bathe method, which
is asymptotically stable with second-order accuracy.
Thanks to its strong dissipation and preferable accu-
racy, the Bathe method has been found to perform
well in many fields [2,4,27]. The three-, and four-sub-
123
H. Zhang et al.
step composite methods [8,32], which are asymptot-
ically stable with higher accuracy, were also devel-
oped adopting the similar idea. Furthermore, to acquire
controllable algorithmic dissipation, the two-sub-step
methods [20,21,26], and the controllable three-sub-
step methods [18,23], were proposed by replacing the
backward difference scheme with a more general for-
mula. However, with the increase in the number of
sub-steps, the number of scalar parameters required to
be designed also increases, so the basic requirements,
including second-order accuracy, unconditional stabil-
ity, controllable algorithmic dissipation, are not enough
to determine these parameters uniquely. Two optimal
sub-families of the controllable three-sub-step method
were proposed in [23], since different conditions are
considered as a supplement.
On this basis, this paper purposes to provide a uni-
versal approach to optimize the parameters of gener-
alized n-sub-step composite method, where n can be
any integer greater than 2, and the trapezoidal rule
is employed in the first n − 1 sub-steps. Two kinds
of optimization goals are considered, producing two
optimal sub-families for different purposes. The first
one intends to achieve higher-order accuracy, under
the premises of unconditional stability and controllable
algorithmic dissipation. The second one is dedicated
to conserving low-frequency behavior, while still pro-
viding controllable high-frequency dissipation. From
linear analysis, the resulting schemes in the first sub-
family can reach up to nth-order accuracy by using n
sub-steps, and the schemes in the second sub-family
exhibit very small algorithmic dissipation in the low-
frequency domain. Most of these schemes are devel-
oped for the first time, and in each sub-family, the
accuracy can be improved by using more sub-steps.
Finally, the proposed methods are applied to solve sev-
eral numerical examples to check the performance.
This paper is organized as follows. The formula-
tions of the n-sub-step composite method are shown in
Sect. 2. The optimization of the parameters is imple-
mented in Sect. 3. The detailed properties of the two
sub-families are discussed in Sect. 4. Numerical exam-
ples are provided in Sect. 5, and conclusions are drawn
in Sect. 6.
On the optimization of n-sub-step composite
2 Formulation
In the literature, the composite methods were mostly
developed to solve the problems in structural dynamics,
as
M ẍ + F (x, ẋ, t) = 0, x (t0 ) = x 0 , ẋ (t0 ) = v 0 (1)
where M is the mass matrix, F collects the damp-
ing force, internal force and external load, x, ẋ and
ẍ are the displacement, velocity and acceleration vec-
tors, respectively, t is the time, and t0 , x 0 and v 0 are the
given initial time, displacement and velocity, respec-
tively. When this method is applied using n sub-steps,
it can be formulated as
 
M ẍ k+2 jγ + F x k+2 jγ , ẋ k+2 jγ , tk + 2 jγ h = 0
(2a)
 
x k+2 jγ = x k+2( j−1)γ + γ h ẋ k+2( j−1)γ + ẋ k+2 jγ
(2b)
 
ẋ k+2 jγ = ẋ k+2( j−1)γ + γ h ẍ k+2( j−1)γ + ẍ k+2 jγ
(2c)
j = 1, 2, 3, · · · , n − 1 (2d)
and
M ẍ k+1 + F (x k+1 , ẋ k+1 , tk + h) = 0 (3a)
⎛ ⎞

n−1
x k+1 = x k + h ⎝ q j ẋ k+2 jγ + qn ẋ k+1 ⎠ (3b)
j=0
⎛ ⎞

n−1
ẋ k+1 = ẋ k + h ⎝ q j ẍ k+2 jγ + qn ẍ k+1 ⎠ (3c)
j=0
where x k ≈ x (tk ) is the numerical solution at step
k, x k+2 jγ ≈ x (tk + 2 jγ h) ( j = 1, 2, · · · , n − 1)
denotes the numerical solution at collocation points,
h is the step size, and γ , q0 , q1 , · · · , qn are the control
parameters. The current step [tk , tk + h] is divided into
n sub-steps: tk , tk + 2γ h , tk + 2γ h, tk + 4γ h ,· · · ,
tk + 2(n − 2)γ h, tk + 2(n − 1)γ h , and [tk +
2(n − 1)γ h, tk + h . In the first n − 1 sub-steps, the
trapezoidal rule is adopted. In the last one, a general
formula containing information about all collocation
points is utilized. The present formulation can reduce
to the ρ∞ -Bathe method [26] when n = 2 and to the
three-sub-step method [18,23] when n = 3.
In this method, because the same form of assump-
tions is used to solve x k+1 and ẋ k+1 , Eqs. (2) and (3)
1941
can be reformulated based on the general first-order
differential equation f ( y, ẏ, t) = 0, as
 
f yk+2 jγ , ẏk+2 jγ , tk + 2 jγ h = 0 (4a)
 
yk+2 jγ = yk+2( j−1)γ + γ h ẏk+2( j−1)γ + ẏk+2 jγ
(4b)
j = 1, 2, 3, · · · , n − 1 (4c)
and
 
f yk+1 , ẏk+1 , tk + h = 0 (5a)
⎛ ⎞

n−1
yk+1 = yk + h ⎝ q j ẏk+2 jγ + qn ẏk+1 ⎠ (5b)
j=0
where {x; ẋ} is replaced by y, and the dynamics equa-
tions can be equivalently formulated as first-order
differential equations by adding the trivial equation
ẋ = ẋ. Equations (4) and (5) present more general for-
mulations for solving first-order and arbitrarily higher-
order differential equations. However, for solving the
second-order dynamic problems, Eqs. (2) and (3) are
still more recommended, since in the equivalent first-
order expressions, the number of implicit equations to
be solved doubles.
From the formulation, the first n − 1 sub-steps
can share the same procedure in a loop, whereas the
last sub-step needs to be implemented separately. The
assumption qn = γ is introduced here, which imposes
that the last sub-step shares the same form of Jacobi
matrix as the first n − 1 sub-steps. This assumption
is particularly useful when applied to linear problems,
since it allows the constant Jacobi matrix to be fac-
torized only once, like in the single-step methods. For
applications, Table 1 shows the computational proce-
dures of the n-sub-step composite method for the gen-
eral first-order differential equation f ( y, ẏ, t) = 0,
where the Newton-Raphson iteration is utilized to solve
the nonlinear equation per sub-step.
Besides, by reorganizing the formulations, the com-
posite method can be regarded as a special case of the
diagonally-implicit Runge–Kutta methods (DIRKs)
with the explicit first-stage. The corresponding Butcher’s
tableau [6] has the form as
123
1942 H. Zhang et al.

Table 1 Computational procedure of the n-sub-step composite method for solving f ( y, ẏ, t) = 0, y(t0 ) = y0

A. Initial calculations
1. From the function f ( y, ẏ, t) and its derivative functions with respect to y and ẏ, as f y and f ẏ , respectively;
2. Initialize t0 , y0 and ẏ0 ;
3. Select the time step size h, the algorithmic parameters γ , q0 , q1 , · · · , qn−1 , the tolerance error ,
and the maximum number of iterations N ;
4. Calculate the constant: a = 1
γh .
B. For each time step
1. The first n − 1 sub-steps
For j = 1, j < n, j + +:
a. Predict yk+2 jγ and ẏk+2 jγ :
i = 0, yk+2 jγ = yk+2( j−1)γ + 2γ h ẏk+2( j−1)γ , ẏk+2 jγ = a( yk+2 jγ − yk+2( j−1)γ ) − ẏk+2( j−1)γ ;
b. Prepare the matrices:
f k+2 jγ = f ( yk+2 jγ , ẏk+2 jγ , tk + 2 jγ h), f y,k+2 jγ = f y ( yk+2 jγ , ẏk+2 jγ , tk + 2 jγ h), f ẏ,k+2 jγ = f ẏ ( yk+2 jγ , ẏk+2 jγ , tk + 2 jγ h);
c. Update yk+2 jγ and ẏk+2 jγ :
i = i + 1,  yk+2 jγ = −( f y,k+2 jγ + a f ẏ,k+2 jγ )−1 f k+2 jγ , yk+2 jγ = yk+2 jγ +  yk+2 jγ , ẏk+2 jγ = ẏk+2 jγ + a yk+2 jγ ;
d. If i < N and | f k+2 jγ | > , go to b; If i ≡ N and | f k+2 jγ | > , abort.
End.
2. The last sub-step
a. Predict yk+1 and ẏk+1 :
n−1
i = 0, yk+1 = yk+2(n−1)γ + (1 − 2(n − 1)γ )h ẏk+2(n−1)γ , ẏk+1 = a( yk+1 − yk − h j=0 q j ẏk+2 jγ );
b. Prepare the matrices:
f k+1 = f ( yk+1 , ẏk+1 , tk + h), f y,k+1 = f y ( yk+1 , ẏk+1 , tk + h), f ẏ,k+1 = f ẏ ( yk+1 , ẏk+1 , tk + h);
c. Update yk+1 and ẏk+1 :
i = i + 1,  yk+1 = −( f y,k+1 + a f ẏ,k+1 )−1 f k+1 , yk+1 = yk+1 +  yk+1 , ẏk+1 = ẏk+1 + a yk+1 ;
d. If i < N and | f k+1 | > , go to b; If i ≡ N and | f k+1 | > , abort.

0 0 0 0 ··· 0 0 order differential equation is discussed, as

2γ γ γ 0 ··· 0 0
.
4γ γ 2γ γ . . 0 0 0 1 x
.. .. .. . . . . . .. ẏ = y, y = (7)
. . . . . .. . −ω2 −2ξ ω ẋ
2(n − 1)γ γ 2γ 2γ · · · γ 0
1 q0 q1 q2 · · · qn−1 γ
Decomposing the coefficient matrix in Eq. (7) yields
q0 q1 q2 · · · qn−1 γ
the simplified first-order equation

3 Optimization 
ẏ = λy, λ = (−ξ ± i 1 − ξ 2 )ω (8)
In linear spectral analysis, owing to the mode superpo-
sition principle, it is common and enough to consider √
the single degree-of-freedom equation where i = −1. When the composite method is
applied, the recursive scheme becomes
ẍ + 2ξ ω ẋ + ω2 x = 0 (6)
where ξ is the damping ratio, and ω is the natural fre-
quency. To simplify the analysis, the equivalent first- yk+1 = A (λh) yk (9)

123
On the optimization of n-sub-step composite 1943

where the amplification factor A is Table 2 Formulas of q j ( j = 0, 1, · · · , n − 1)

⎛ ⎞ n q j ( j = 0, 1, · · · , n − 1)

n−1  j
1+γz
A (z) = (1 − qn z)−1 ⎝1 + z qj ⎠, γ a1 a2
1−γz 2 q0 = + −
j=0 2 2 2γ
γ a1 a2
z = λh (10) q1 = + +
2 2 2γ
3γ a1 a2 a3
Since qn = γ is assumed in Sect. 2, Eq. (10) is updated 3 q0 = + − +
4 4 4γ 4γ 2
as a1 a3
q1 =γ + −
n−1   2 2γ 2
(1 − γ z)n−1 + z j=0 q j (1 + γ z) j (1 − γ z)n− j−1
A (z) = γ a1 a2 a3
(1 − γ z)n q2 = + + +
(11) 4 4 4γ 4γ 2
7γ a1 a2 a3 a4
1 + a1 z + a2 z 2 + · · · + an z n 4 q0 = + − + −
= 8 8 8γ 8γ 2 8γ 3
(1 − γ z)n
11γ 3a1 a2 a3 3a4
q1 = + − − +
8 8 8γ 8γ 2 8γ 3
where the coefficient of z p ( p = 1, 2, · · · , n) is repre-
5γ 3a1 a2 a3 3a4
sented by a p ( p = 1, 2, · · · , n), expressed as q2 = + + − 2
−
8 8 8γ 8γ 8γ 3
  γ a1 a2 a3 a4
p q3 = + + + +
ap = (−γ ) p 8 8 8γ 8γ 2 8γ 3
n−1
⎛ ⎞ 5 q0 =
15γ
+
a1
−
a2
+
a3
−
a4
+
a5

n−1 min{
j, p−1} 16 16 16γ 16γ 2 16γ 3 16γ 4
+ γ p−1 ⎝q j P(m, j, p, n)⎠ , 13γ a1 a2 a4 a5
q1 = + − + −
j=0 m=max{0, p+ j−n} 8 4 8γ 8γ 3 4γ 4
  
p−m−1 m p−m−1 3a1 a3 3a5
P(m, j, p, n) = (−1) , q2 =γ + − +
j n− j −1 8 8γ 2 8γ 4
3γ a1 a2 a4 a5
p = 1, 2, · · · , n − 1 (12) q3 = + + − −
8 4 8γ 8γ 3 4γ 4
γ a1 a2 a3 a4 a5
and q4 = + + + + +
16 16 16γ 16γ 2 16γ 3 16γ 4
n−1 
 
an = γ n−1 (−1)n− j−1 q j (13)
j=0
the formulas of q j expressed by a p and γ for the cases
n = 2, 3, 4, 5.
For example, n = 5 follows

a1 = −4γ + q0 + q1 + q2 + q3 + q4 (14a)
a2 = 6γ + γ (−4q0 − 2q1 + 2q3 + 4q4 )
2
(14b)
a3 = −4γ + γ (6q0 − 2q2 + 6q4 )
3 2
(14c) 3.1 Higher-order schemes

a4 = γ 4 + γ 3 (−4q0 + 2q1 − 2q3 + 4q4 ) (14d)

a5 = γ (q0 − q1 + q2 − q3 + q4 )
4
(14e)
Consequently, the parameters under analysis change
from q j ( j = 0, 1, · · · , n − 1) and γ , to a p ( p = 1, 2, · · · , n)
and γ in the following. When a p and γ are given,
the parameters q j can be obtained uniquely by solv-
ing Eqs. (12) and (13). For applications, Table 2 shows
A numerical method is naturally expected to be as accu-
rate as possible, so the higher-order schemes are con-
sidered first. From the scheme of Eq. (9), the composite
method uses the amplification factor A, rewritten as
1 + a1 z + a2 z 2 + · · · + an z n
A (z) = (15)
(1 − γ z)n

123
1944 H. Zhang et al.

to approximate the exact amplification factor  which are the numerator and denominator of A(z) in
Eq. (15), respectively, |A(z)| ≤ 1 is equivalent to
1 1
 (z) = ez = 1 + z + z 2 + z 3 + · · · (16) N (iτ )N (−iτ )
2 6 |A(z)|2 = A(iτ )A(−iτ ) = ≤1 (21)
D(iτ )D(−iτ )
Hence the local truncation error σ can be defined as
Then the condition for unconditional stability can be
  transformed into
σ = yk+1 − y (tk+1 ) = A(z) − Â(z) y(tk ) (17)
S(τ ) = D(iτ )D(−iτ ) − N (iτ )N (−iτ )
  n  
If σ = O z s+1
, the method is said to be sth-order
= c2 j τ 2 j ≥ 0 for τ ≥ 0 (22)
accurate, which requires that up to sth derivatives of A
j=0
at z = 0 are all equal to 1, that is
where the function S(τ ) is introduced, and the coeffi-
A(0) = A(1) (0) = A(2) (0) = · · · = A(s) (0) = 1 (18) cients c2 j ( j = 0, 1, 2, · · · , n) are expressed as
 
To satisfy Eq. (18), a p ( p = 1, 2, · · · , n) can be solved j 2j
c2 j = γ
as n
 j}
min{n,2
 
(1) + (−1) j+1
(−1)m am a2 j−m
A (0) = 1 ⇒ a1 = 1 − nγ (19a) m=max{0,2 j−n}
1 n(n − 1)
A(2) (0) = 1 ⇒ a2 = − nγ + γ2 (19b) (23)
2 2
A(3) (0) = 1 ⇒ (19c) in which a0 is set to 1. By Eq. (22), the bounds on γ
1 n n(n − 1) 2 n(n − 1)(n − 2) 3 of the cases n = 2, 3, 4, 5 are provided in Table 3 .
a3 = − γ + γ − γ It follows that, with s = n, the allowable range of γ
6 2 2 6
··· (19d) narrows as n increases and, in some cases, the n-sub-
    step method can achieve (n + 1)th-order accuracy with
 (−1) j j
s
A(s) (0) = 1 ⇒ as = γj (19e) a fixed γ .
(s − j)! n Besides, algorithmic dissipation is also a desirable
j=0
property for a time integration method, to filter out the
inaccurate high-frequency content. Generally, it is mea-
Therefore, if all a p ( p = 1, 2, · · · , n) follow the rela-
sured by the spectral radius ρ∞ at high-frequency limit,
tionships in Eq. (19), this method can achieve nth-order
that is
accuracy, and then γ becomes the only free parameter
to control the stability.
A time integration method is said to be uncondi- |A(z)| → ρ∞ as |z| = ωh → +∞, ρ∞ ∈ [0, 1]
tionally stable if |A(z)| (24)
 ≤ 1 for all R(z) ≤ 0 where
z = λh = (−ξ ± i 1 − ξ 2 )ωh. According to Ref.
[19], the bounds on γ can be given by considering the and it gets stronger with a smaller ρ∞ . With A(z) from
stability on the imaginary axis (ξ = 0), which can result Eq. (15), Eq. (24) can be satisfied if
in the unconditional stability when the accuracy order ⎛ ⎞2
s = n in the DIRKs. Therefore, let z = ±iτ where n  j  j 
(−1)
τ = ωh is a real number, and an2 = ⎝ γ j ⎠ = ρ∞ γ
2 2n
(25)
(n − j)! n
j=0

N (z) = 1 + a1 z + a2 z 2 + · · · + an z n (20a)
which can be used to solve γ for a given ρ∞ . Table 4
D(z) = (1 − γ z) n
(20b) shows the solutions of γ for several specific ρ∞ in

123
On the optimization of n-sub-step composite 1945

Table 3 Bounds on γ for unconditional stability (s is the accuracy order) in the higher-order schemes
n s Bounds on γ

2 2 [0.250000000000000, +∞]
3 0.788675134594813
3 3 [0.333333333333333, 1.068579021301628]
4 1.068579021301628
4 4 [0.394337567297396, 1.280579761275305]
5 –

5 5 [0.246505193142435, 0.361803398875471] [0.420782512765729, 0.473268391258294]
6 0.473268391258294

the cases n = 2, 3, 4, 5. Note that Eq. (25) has multi- 1 n(n − 1) 2

a2 = − nγ + γ (26b)
ple solutions; the smallest one that meets the require- 2 2
ment of unconditional stability, as shown in Table 3, is
selected. and controllable algorithmic dissipation, achieved by
So far, the unconditionally stable higher-order accu-
rate schemes with controllable algorithmic dissipation an2 = ρ∞ γ
2 2n
(27)
have been developed, whose parameter γ can be solved
for a given ρ∞ by Eq. (25), a p ( p = 1, 2, · · · , n) In addition, unconditional stability also needs to be sat-
are determined by γ as shown in Eq. (19), and then isfied, which will be checked last.
q j ( j = 0, 1, 2, · · · , n − 1) can be obtained by solv- To conserve the energy as much as possible, the
ing Eqs. (12) and (13). These information for the cases spectral radius ρ = |A(z)| should be as close to 1 as
n = 2, 3, 4, 5 are shown in Tables 2, 3 and 4. The possible over the low-frequency range. For the special
special case n = 2 is identical to the ρ∞ -Bathe method case ρ∞ = 1, ρ should remain 1 in the whole fre-
[26], whereas the other cases are presented here for quency domain. For other cases 0 ≤ ρ∞ < 1, the
the first time. In addition, the accuracy and algorithmic departure of ρ from unit value should be as slow as
dissipation are discussed in more detail in Sect. 4. possible from ρ(0) = 1. Considering the conservative
system (ξ = 0), this purpose can be realized by making
the function S(τ ), defined in Eq. (22), as smooth as pos-
3.2 Conserving schemes sible. It follows that S(0) = S (1) (0) = S (2) (0) = · · · =
S (m) (0) = 0, where S (m) (0) is the mth-order derivative
An original intention of the composite methods was to of S(τ ) at τ = 0, and m should be as large as possible.
conserve the energy of the system [2], which explains As S(τ ) is a linear polynomial, the condition transforms
why the trapezoidal rule is utilized in most sub-steps. into its coefficients c2 j ( j = 0, 1, 2, · · · , m) = 0. To
Existing two- and three-sub-step methods [18,22,26] clarify, c2 j ( j = 0, 1, 2, · · · , n) are enumerated as
really show preferable energy-conserving characteris-
tic over other single- and multi-step methods. In this c0 = 0 (28a)
work, a simple and general approach to determine the c2 = nγ 2
− a12
+ 2a0 a2 (28b)
parameters, which enable the n-sub-step composite
n(n − 1) 4
method to conserve as much low-frequency content as c4 = γ − a22 + 2a1 a3 − 2a0 a4 (28c)
2
possible, is proposed.
n(n − 1)(n − 2) 6
First of all, to be competitive, the method needs to c6 = γ − a32 + 2a2 a4 − 2a1 a5
6
have some basically useful properties, including at least
+ 2a0 a6 (28d)
second-order accuracy, which requires
··· (28e)
a1 = 1 − nγ (26a) c2n−2 = nγ 2n−2 − an−1
2
+ 2an−2 an (28f)

123
1946 H. Zhang et al.

Table 4 γ for controllable algorithmic dissipation in the higher-order schemes

ρ∞ n=2 n=3 n=4 n=5

0.0 0.292893218813452 0.435866521508460 0.572816062482135 0.278053841136450

0.1 0.287089056989371 0.421486815409409 0.548366644975830 0.274141306031868
0.2 0.281754163448146 0.408500789512922 0.526386456842386 0.270459886774582
0.3 0.276820321671636 0.396647209121134 0.506330118970782 0.266978043925651
0.4 0.272233289109874 0.385731000460835 0.487797474812348 0.263670231711606
0.5 0.267949192431123 0.375602225015285 0.470480577621677 0.260515416607055
0.6 0.263932022500210 0.366142810103347 0.454130785036529 0.257496029856675
0.7 0.260151847569038 0.357257811967234 0.438536189902193 0.254597208170133
0.8 0.256583509747431 0.348869453074869 0.423503766067179 0.251806231183850
0.9 0.253205655191036 0.340912922771929 0.408841866120699 0.249112096529630
1.0 0.250000000000000 0.333333333333333 0.394337567297407 0.246505193142820

c2n = γ 2n − an2 (28g) all schemes, so these methods can be said to possess
unconditional stability for linear problems. Other prop-
From Eqs. (26) and (27), we can obtain c2 = 0 and erties are discussed in Sect. 4.
c2n = (1 − ρ∞ 2 )γ 2n ≥ 0, respectively. For the case

n = 2, the conditions on accuracy and algorithmic dis-

sipation are enough to determine all parameters, result-
ing again in the ρ∞ -Bathe method [26]. For other cases
with n > 2, the n − 2 remaining parameters, a3 , a4 ,
· · · , an−1 and γ , are obtained by solving the equa-
tions c4 = c6 = · · · = c2n−2 = 0. The values of
these parameters for the cases n = 3, 4, 5 are shown in
Table 5, where the set with γ close to 2n 1
is selected,
which requires an = ρ∞ γ n .
Then all parameters of the conserving schemes have
been given by combining Eqs. (12), (13), (26), (27) and
c4 = c6 = · · · = c2n−2 = 0. The resulting scheme of
n = 3 is equivalent to the first sub-family of the three-
sub-step method proposed in [23]; the other cases are
presented here for the first time.
In particular, when ρ∞ = 1, the resulting scheme is
a n-sub-step method with the trapezoidal rule in all sub-
steps, which is supposed to be unconditionally stable
in the linear analysis. Empirically, the algorithmic dis-
sipation is acquired by reducing the spectral radius ρ,
so the dissipative schemes are likely to be also uncon-
ditionally stable, and even present more robust stabil-
ity. For the undamped case (ξ = 0), the stability can
be guaranteed since S(τ ) = (1 − ρ∞ 2 )γ 2n τ 2n ≥ 0;
for other cases, the stability conditions of the schemes
listed in Table 5 are checked one by one by consid-
ering ξ ∈ (0, 1] and τ ∈ [0, 10000] numerically.
As expected, ρ ≤ 1 is satisfied at every point in
4 Properties
Two sub-families of the n-sub-step composite method
have been presented for different purposes. To iden-
tify them, the higher-order schemes are referred to as
MSSTH(n), and the conserving schemes are MSSTC(n),
where MSST means the multi-sub-step composite
method which employs the trapezoidal rule in all sub-
steps except the last one, H and C are utilized to dis-
tinguish the two sub-families, and n is the number of
sub-steps.
In this section, the representative methods in the lit-
erature, including the single-step generalized-α method
[9] (G-α) and the linear two-step method [33] (LTS) are
also considered for comparison. As the employed meth-
ods are all implicit, their computational cost is mainly
spent on the iterative calculation when used for non-
linear problems, or the matrix factorization for linear
problems. The vector operations brought by the recur-
sive scheme of the method itself is generally considered
to have little effect on overall efficiency. Therefore, G-α
and LTS are recognized as having equivalent efficiency
if the same step size is used. As the composite methods
implement a single-step or multi-step scheme in each
sub-step, they have the equivalent efficiency to G-α and
LTS, if their required number of sub-steps is equal to
the number of steps required by G-α and LTS. For this

123
On the optimization of n-sub-step composite 1947

Table 5 a p ( p = 3, 4, · · · , n − 1) and γ for controllable algorithmic dissipation in the conserving schemes

ρ∞ n = 3 n=4 n=5
γ a3 γ a3 a4 γ

0.0 0.180425306429398 0.00453529185986996 0.131378736730466 0.00763819606391975 0.000257160742971488 0.103557108920215

0.1 0.178619458204658 0.00494493283913114 0.130548620946472 0.00793598250555122 0.000286872844754286 0.103095631511675
0.2 0.176945806618224 0.00532986673141648 0.129777583818848 0.00821506063336095 0.000314958374958848 0.102666675025093
0.3 0.175385515842846 0.00569340893903897 0.129057207257355 0.00847799948547669 0.000341631205811411 0.102265594492185
0.4 0.173923607877197 0.00603821324069812 0.128380804919945 0.00872684827484184 0.000367062789628561 0.101888703879882
0.5 0.172547961422089 0.00636644119939074 0.127742970556848 0.00896327074168002 0.000391393000239752 0.101533025147874
0.6 0.171248618590691 0.00667987987582935 0.127139265902084 0.00918863822587218 0.000414737666914113 0.101196115073181
0.7 0.170017291772476 0.00698002584685393 0.126565999083137 0.00940409590940791 0.000437193952981663 0.100875942445807
0.8 0.168847004679168 0.00726814657645952 0.126020063586496 0.00961061107396470 0.000458844285976519 0.100570798918745
0.9 0.167731825756887 0.00754532615554921 0.125498818830422 0.00980900897660744 0.000479759288032603 0.100279232954742
1.0 0.166666666666667 0.00781250000000000 0.125000000000000 0.01000000000000000 0.000500000000000000 0.100000000000000

Fig. 1 Percentage amplitude decay for MSSTH(2,3,4,5), G-α and LTS

reason, to compare the properties under the close com-
putational costs, the same h/n, where n is the number
of sub-steps in the composite methods, and n = 1 for
the G-α and LTS, is used in these methods.
As discussed in Sect. 3, MSSTH(n) has nth-order
accuracy under the premises of unconditional stability
and controllable algorithmic dissipation. Figures 1 and
2 display the percentage amplitude decay (AD(%)) and
period elongation (PE(%)) respectively, of which the
definition can refer to [34], of MSSTH(2,3,4,5), G-α
and LTS, considering the undamped case (ξ = 0). The
abscissa is set as τ/n to compare these methods under
the close computational costs.
The results illustrate that the amplitude and period
accuracy cannot be improved simultaneously as the
order of accuracy increases in MSSTH(n). In terms
of amplitude, with a small ρ∞ , the ρ∞ -Bathe method
(the same as MSSTH(2)) is the most accurate, and when
0.4 < ρ∞ ≤ 1, LTS shows smaller dissipation error,
followed by the G-α and the ρ∞ -Bathe method. From
Fig. 2, MSSTH(3,4,5) have smaller period error than
the second-order methods, and MSSTH(5) is the best
among them.
In the same way, the percentage amplitude decay
and period elongation of MSSTC(2,3,4,5), G-α and
LTS for the undamped case are shown in Figs. 3 and 4,

123
1948 H. Zhang et al.

Fig. 2 Percentage period elongation for MSSTH(2,3,4,5), G-α and LTS

Fig. 3 Percentage amplitude decay for MSSTC(2,3,4,5), G-α and LTS

respectively. It can be observed that under the similar
efficiency, MSSTC(n) presents higher amplitude and
period accuracy with a larger n. The gap is more obvi-
ous as ρ∞ decreases, and when ρ∞ = 1, all the schemes
have the same properties as the trapezoidal rule. Both
G-α and LTS are less accurate than MSSTC(3,4,5) in
the low-frequency range.
Besides, with the same n, MSSTH(n) and MSSTC(n)
are compared in Figs. 5, 6, 7, 8, 9, 10, 11, 12 and 13,
where Figs. 5, 6 and 7 show the spectral radius (SR)
of the cases n = 3, 4, 5, respectively, Figs. 8, 9 and 10
show the percentage amplitude decay, Figs. 11, 12 and
13 show the percentage period elongation, all consider-
ing the undamped case. The generalized Padé approxi-
mation [14,15], referred to as Padé(n), is also employed
for comparison. It is known as the most accurate ratio-
nal approximation of ez by using
(1 − ρ∞ )Pn−1,n (z) + 2ρ∞ Pn,n (z)
A(z) = (29)
(1 − ρ∞ )Q n−1,n (z) + 2ρ∞ Q n,n (z)

123
On the optimization of n-sub-step composite 1949

Fig. 4 Percentage period elongation for MSSTC(2,3,4,5), G-α and LTS

Fig. 5 Spectral radius for n = 3

where As expected, Figs. 5, 6 and 7 demonstrate that

MSSTC(n) preserves wider low-frequency range, fol-

i
i!( j + i − p)! z p lowed by Padé(n), and MSSTH(n). Note that MSSTH(n)
Pi, j (z) = (30a) with ρ∞ = 1 exhibits mild algorithmic dissipation
(i − p)!( j + i)! p!
p=0 in the medium frequency range, so these schemes are

i
j!( j + i − p)! z p not recommended if all frequencies are requested. Fig-
Q i, j (z) = (−1) p (30b) ures 8, 9 and 10 also show that MSSTC(n) has the
( j − p)!( j + i)! p!
p=0 smallest amplitude dissipation in the low-frequency
content. The amplitude decay ratio of MSSTC(5) is
Padé(n) has (2n − 1)th-order accuracy if 0 ≤ ρ∞ < 1 very close to 0 over τ ∈ [0, 2]. In terms of period accu-
and (2n)th-order accuracy if ρ∞ = 1. racy, Figs. 11, 12 and 13 show that Padé(n) is the most

123
1950 H. Zhang et al.

Fig. 6 Spectral radius for n = 4

Fig. 7 Spectral radius for n = 5

accurate, followed by MSSTH(n), and MSSTC(n), 5 Numerical examples

consistent with the sequence of the accuracy order.
From the comparison, MSSTC(n) performs really
good at conserving the low-frequency content, and its
overall accuracy can be improved by using more sub-
steps. MSSTH(n) shows higher period accuracy than
the second-order methods, whereas its dissipation error
is larger in the low-frequency content.
To validate the performance, several numerical exam-
ples are solved in this section. As the spectral analysis
has revealed the properties based on the linear model,
this section focuses more on the application and dis-
cussion for nonlinear systems.

123
On the optimization of n-sub-step composite
Fig. 8 Percentage amplitude decay for n = 3
Fig. 9 Percentage amplitude decay for n = 4
5.1 Single degree-of-freedom examples
Firstly, two single degree-of-freedom examples, includ-
ing a simple linear example and the nonlinear van
der Pol’s equation, are solved to check the conver-
gence rate. The ρ∞ -Bathe method, MSSTC(3,4,5) and
MSSTH(3,4,5) with ρ∞ = 0.6 is employed.
Linear example The linear equation of motion
ẍ + 4x = 0, x(0) = 1, ẋ(0) = 1 (31)
1951
is considered, and the absolute errors of the displace-
ment xk , velocity ẋk , and acceleration ẍk versus h at
t = 10 are plotted in Fig. 14.
The results are consistent with the accuracy order.
That is, MSSTC(n) and MSSTH(n) respectively present
second-order and nth-order convergence rate. As a
result, the higher-order MSSTH(n) enjoys significant
accuracy advantage over the second-order methods.
However, when h decreases from 10−2 , it seems that
MSSTH(5) cannot maintain fifth-order accuracy. This
is because when h is small enough, all effective num-
123
1952 H. Zhang et al.

Fig. 10 Percentage amplitude decay for n = 5

Fig. 11 Percentage period elongation for n = 3

bers stored in the computer are exactly precise, so if h is solved, where  is an adjustable parameter. For the
continues to decrease, the accumulated rounding error cases  = 0.01, 0.001, 0.0001, the absolute errors of
can greatly spoil the numerical precision [31]. x1,k and x2,k at t = 1 versus h are plotted in Fig. 15,
Van der Pol’s equation The van der Pol’s equation [19] where the reference solution is obtained by the ρ∞ -
Bathe method with h = 10−7 .
From Fig. 15, in most cases, the second- and nth-
ẋ1 = x2 , ẋ2 =  −1 ((1 − x12 )x2 − x1 ) order convergence rate can be observed from errors of
MSSTC(n) and MSSTH(n), respectively, but for the
2 10 292 2 15266 3
x1 (0) = 2, x2 (0) = − +  −  +  stiffer case of  = 0.0001, MSSTH(3) and MSSTH(5)
3 81 2187 59049 show obvious order reduction in both x1 and x2 . It indi-
(32)

123
On the optimization of n-sub-step composite
Fig. 12 Percentage period elongation for n = 4
Fig. 13 Percentage period elongation for n = 5
cates that the accuracy order also depends on the prob-
lem to be solved when applied to nonlinear systems.
The order reduction also occurs in other higher-order
DIRKs when used for nonlinear problems, see Ref.
[19]. Nevertheless, MSSTH(n) still shows significant
accuracy advantage over the second-order MSSTC(n)
with a small step size.
1953
5.2 Multiple degrees-of-freedom examples
In this subsection, some illustrative examples are
solved by using the ρ∞ -Bathe method, MSSTC(3,4,5),
MSSTH(3,4,5), G-α and LTS. In these methods, the
parameter ρ∞ is set as 0 uniformly, and the same h/n
is used for comparison under close computational costs.
The reference solutions are obtained by the ρ∞ -Bathe
method with an extremely small time step.
123
1954 H. Zhang et al.

Fig. 14 Convergence rates for the single degree-of-freedom linear example

Fig. 15 Convergence rates for the van der Pol’s equation

one end and with a mass at the free end, is simulated.

Its motion equation can be written as

m r̈ + f (r ) − m(L 0 + r )θ̇ 2 − mg cos θ = 0 (33a)

m(2ṙ θ̇ + g sin θ )
m θ̈ + =0 (33b)
L0 + r

where f (r ) denotes the elastic force of the spring and

other system parameters are assumed as m = 1 kg,
L 0 = 0.5 m, g = 9.81 m/s2 . Three kinds of constitutive
relations, as

Fig. 16 Spring-pendulum model

f (r ) = kr (34a)
f (r ) = kr 3
(34b)
Spring-pendulum model As shown in Fig. 16, the
spring-pendulum model, where the spring is fixed at f (r ) = k tanh r (34c)

123
On the optimization of n-sub-step composite 1955

Fig. 17 Numerical results

of the spring-pendulum
model ( f (r ) = kr ,
k = 98.1 N/m)

Fig. 18 Numerical results

of the spring-pendulum
model ( f (r ) = kr 3 ,
k = 98.1 N/m)

123
1956 H. Zhang et al.

Fig. 19 Numerical results

of the spring-pendulum
model ( f (r ) = k tanh r ,
k = 98.1 N/m)

where k = 98.1 N/m, are considered. The initial con- Moreover, to check the algorithmic dissipation, the
ditions are set as stiff case, where f (r ) = kr (k = 98.1 × 1010 N/m),
is also simulated with h/n = 0.01 s. The numerical
π results of E − E 0 , r and θ are plotted in Fig. 20. The
r0 = 0 m, ṙ0 = 1 m/s, θ0 = rad, θ̇0 rad/s (35) results of r indicate that all employed schemes with
4
ρ∞ = 0 can effectively filter out the stiff compo-
nent in the first few steps. After the initial decaying,
Let h/n = 0.01 s; the numerical solutions of E − E 0
MSSTC(3,4,5) can still preserve the remaining energy
(E denotes the system energy and E 0 is the initial
in the following simulation.
value), r and θ for the three cases are summarized in
Slider-pendulum model The slider-pendulum model,
Figs. 17, 18 and 19. From the curves of E − E 0 , it
shown in Fig. 21, is considered in this case. The slider is
can be observed that MSSTC(3,4,5) can almost pre-
constrained by the spring, and one end of the pendulum
serve the numerical energy from decaying in all cases,
is hinged to the center of mass of the slider. The motion
despite the oscillations. MSSTH(5) can preserve more
is described by the differential-algebraic equations
energy than the Bathe method, while G-α, LTS, and
MSSTH(3,4) show obvious energy-decaying. From the
numerical results of r and θ , one can see that with m 1 ẍ1 + kx1 = −λ1 (36a)
the step size, the numerical solutions of these meth-
m 2 ẍ2 = λ1 (36b)
ods have clearly deviated from the reference solution
after a period of simulation. Among these methods, m 2 ÿ2 = λ2 − m 2 g (36c)
MSSTH(5) predicts the closest solutions to the refer- L L
J2 θ̈ = − λ1 cos θ − λ2 sin θ (36d)
ence ones, and G-α shows the largest errors. In addi- 2 2
tion, MSSTC(3,4,5) exhibit good amplitude accuracy L
x2 − x1 = sin θ (36e)
thanks to their energy-preserving characteristic. These 2
conclusions are all consistent with the results from lin- L
y2 = − cos θ (36f)
ear analysis. 2

123
On the optimization of n-sub-step composite 1957

Fig. 20 Numerical results

of the spring-pendulum
model ( f (r ) = kr ,
k = 98.1 × 1010 N/m)

The system parameters are m 1 = m 2 = 1 kg, L =

1 m, J2 = 12 1
kg · m2 , g = 9.81 m/s2 , k = 1 N/m
10
and 10 N/m respectively for the compliant and stiff
systems. The slider is excited by the initial horizontal
velocity 1 m/s.
By using h/n = 0.01 s, the numerical solutions of
E − E 0 , x1 and θ for the compliant and stiff cases
are shown in Figs. 22 and 23, respectively. From the
results of x1 and θ , these methods all perform well in
Fig. 21 Slider-pendulum model
terms of accuracy and algorithmic dissipation. How-
ever, the numerical energies of MSSTH(4) show a
slightly upward trend in the stiff case, so this method
cannot give stable results for the problem. MSSTH(4) is not recommended for these problems due
As already discussed in several papers [5,30], the to its poorer stability.
unconditional stability of a time integration method N -degree-of-freedom mass-spring system The N -degree-
derived from linear analysis cannot be guaranteed when of-freedom mass-spring system [23], shown in Fig. 24,
they are applied to nonlinear problems. For nonlin- is considered to check the computational efficiency.
ear problems, the stability of a method depends not The system parameters are set as
only on its recursive scheme, but also on the problem
itself. Therefore, it is hard to give a definite conclusion
about the stability of a method for general problems. m i = 1 kg, f i = sin t N, i = 1, 2, · · · , N (37a)
From the numerical results, all employed methods, ⎧
except MSSTH(4), provide stable results when solv- ⎨ 105 N/m, i = 1
ki =  
ing stiff problems and differential-algebraic equations, ⎩ 105 1 − 2(xi − xi−1 )2 N/m, 2 ≤ i ≤ N
so they can be said to have relatively strong stability.
(37b)

123
1958 H. Zhang et al.

Fig. 22 Numerical results

of the slider-pendulum
model (k = 1 N/m)

Fig. 23 Numerical results

of the slider-pendulum
model (k = 1010 N/m)

123
On the optimization of n-sub-step composite
Fig. 24 Mass-spring model
With zero initial conditions, three cases, N =
500, 1000 and 1500, are simulated by these methods
using h/n = 0.01 s. Figure 25 shows the numerical
solutions of x N . It follows that with the step size, all
methods can provide reliable results. The CPU time
and total number of iterations required by these meth-
ods in the simulation of [0, 30 s] are summarized in
Table 6. With h/n = 0.01 s, these methods need to
proceed 3000 steps (sub-steps for the composite meth-
ods) in the whole simulation. Table 6 shows that in
addition to MSSTH(4,5), other methods only require
one iteration per step/sub-step, so their computational
costs are almost equal to each other. One can also see
that the required CPU time is approximately propor-
tional to the number of iterations. MSSTH(4,5), espe-
cially MSSTH(4), take slightly longer time than other
methods.
To check the generality of this conclusion, the
required total number of iterations in the above spring-
Fig. 25 Computed x N of
the mass-spring model
1959
pendulum and slider-pendulum examples are also listed
in Table 7. In the two examples, h/n = 0.01 s is
adopted, and the simulation of [0, 30 s] is also per-
formed. The results indicate that the total numbers
of iterations required by the second-order methods
are very close in all cases. Although the higher-order
methods need more iterations sometimes, the increased
numbers, especially in MSSTH(3,5), are not very large
in most cases. Therefore, it is reasonable to say that
these methods with the same h/n have similar effi-
ciency for nonlinear problems, and the above compar-
isons in terms of properties are conducted under the
close computational costs.
Overall, the numerical examples in this section
demonstrate that when applied to nonlinear problems,
the proposed methods can still take advantage of their
properties, including the energy-conserving character-
istic of MSSTC(n), the high-accuracy of MSSTH(n),
and the strong dissipation ability of both sub-families.
However, MSSTH(n) shows reduced order and energy
instability in some examples. From the presented solu-
tions, MSSTH(5) is more recommended in the higher-
order sub-family because of its high accuracy and
robust stability, whereas MSSTH(4) is not so prefer-
123
1960 H. Zhang et al.

Table 6 CPU time and total number of iterations required by these methods in the mass-spring example
Method N = 500 N = 1000 N = 1500
CPU time (s) Number of iterations CPU time (s) Number of iterations CPU time (s) Number of iterations

G-α 34.8852 3000 155.0207 3000 364.3084 3000

LTS 34.1053 3000 164.0188 3000 361.6325 3000
Bathe 33.6191 3000 162.6447 3000 367.3689 3000
MSSTC(3) 34.2762 3000 164.4059 3000 363.7048 3000
MSSTC(4) 32.6436 3000 164.3387 3000 365.8208 3000
MSSTC(5) 33.3455 3000 162.9475 3000 361.8247 3000
MSSTH(3) 38.1994 3000 162.6028 3000 360.1350 3000
MSSTH(4) 52.6115 3988 165.4999 3012 383.2466 3306
MSSTH(5) 47.8183 3289 163.5020 3000 369.4852 3008

Table 7 Total number of iterations required by these methods in the spring-pendulum and slider-pendulum example
Method Spring-pendulum example Slider-pendulum example
f (r ) = kr f (r ) = kr 3 f (r ) = k tanh r f (r ) = kr (Stiff case) Compliant case Stiff case

G-α 5999 5999 5997 5925 3161 5980

LTS 5999 6000 5997 5923 3164 5978
Bathe 5998 5997 5997 5902 3285 6001
MSSTC(3) 5999 5998 5998 5890 3000 5978
MSSTC(4) 5999 5998 6000 5893 3000 6003
MSSTC(5) 5999 5999 5997 5891 3000 5992
MSSTH(3) 6000 6001 6000 5989 5751 6007
MSSTH(4) 8405 8286 8355 6006 5973 6663
MSSTH(5) 6704 6583 6676 5998 5828 6139

able, since it shows energy-instability and needs more
iterations in some examples.
6 Conclusions
In this work, the n-sub-step composite method (n ≥ 2),
which employs the trapezoidal rule in the first n − 1
sub-steps and a general formula in the last one, is dis-
cussed. By optimizing the parameters, the two sub-
families, named MSSTC(n) and MSSTH(n), are devel-
oped, respectively for the energy-conserving and high-
accuracy purposes. From linear analysis, MSSTC(n)
and MSSTH(n) are second-order and nth-order accu-
rate, respectively, and they can both achieve uncondi-
tional stability with controllable algorithmic dissipa-
tion. In MSSTC(n), the purpose of energy-conserving
is realized by maximizing the spectral radius in the
low-frequency range.
A general approach of parameter optimization, suit-
able for all schemes with n ≥ 2, is proposed; in
this work, the cases n = 2, 3, 4, 5 are discussed in
detail. When n = 2, both sub-families reduce to the
ρ∞ -Bathe method. As n increases, MSSTC(n) shows
higher amplitude and period accuracy; its amplitude
accuracy is even higher than that of the (2n − 1)th-
order Padé(n) approximation. MSSTH(3,4,5) exhibits
lower period errors than the second-order methods, but
their dissipation errors are larger.
The proposed methods are checked on several illus-
trative examples. The numerical results are mostly con-
sistent with the conclusions from linear analysis. That
is, MSSTC(n) can conserve the energy corresponding
to the low-frequency content, and MSSTH(n) shows
higher-order convergence rate for linear and nonlin-
ear equations. However, in the nonlinear examples,
some unexpected situations, such as order reduction

123
On the optimization of n-sub-step composite
and energy instability, emerged in MSSTH(n). In this
sub-family, MSSTH(5) is more recommended thanks to
its high-accuracy and robust stability, and MSSTH(4) is
not so preferable, since it shows energy instability and
lower efficiency in these examples. However, these con-
clusions about nonlinear problems are obtained from
the existing numerical results. The theoretical analysis
is still desired in the future.
Acknowledgements The first and second authors acknowl-
edge the financial support by the China Scholarship Council.
Funding Open access funding provided by Politecnico di
Milano within the CRUI-CARE Agreement.
Compliance with ethical standards
Conflict of interest The authors declare that they have no con-
flict of interest.
Open Access This article is licensed under a Creative Com-
mons Attribution 4.0 International License, which permits use,
sharing, adaptation, distribution and reproduction in any medium
or format, as long as you give appropriate credit to the original
author(s) and the source, provide a link to the Creative Com-
mons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s
Creative Commons licence, unless indicated otherwise in a credit
line to the material. If material is not included in the article’s Cre-
ative Commons licence and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need
to obtain permission directly from the copyright holder. To view
a copy of this licence, visit http://creativecommons.org/licenses/
by/4.0/.
References
1. Bank, R.E., Coughran, W.M., Fichtner, W., Grosse, E.H.,
Rose, D.J., Smith, R.K.: Transient simulation of silicon
devices and circuits. IEEE Trans. Comput. Aided Des.
Integr. Circuits Syst. 4(4), 436–451 (1985)
2. Bathe, K.J.: Conserving energy and momentum in nonlinear
dynamics: a simple implicit time integration scheme. Com-
put. Struct. 85(7–8), 437–445 (2007)
3. Bathe, K.J., Baig, M.M.I.: On a composite implicit time inte-
gration procedure for nonlinear dynamics. Comput. Struct.
83(31–32), 2513–2524 (2005)
4. Bathe, K.J., Noh, G.: Insight into an implicit time integration
scheme for structural dynamics. Comput. Struct. 98, 1–6
(2012)
5. Belytschko, T., Schoeberle, D.: On the unconditional sta-
bility of an implicit algorithm for nonlinear structural
dynamics. J. Appl. Mech. (1975). https://doi.org/10.1115/
1.3423721
6. Butcher, J.C.: Implicit Runge–Kutta processes. Math. Com-
put. 18(85), 50–64 (1964)
1961
7. Butcher, J.C.: Numerical Methods for Ordinary Differential
Equations. John Wiley, New Jersey (2016)
8. Chandra, Y., Zhou, Y., Stanciulescu, I., Eason, T.,
Spottswood, S.: A robust composite time integration scheme
for snap-through problems. Comput. Mech. 55(5), 1041–
1056 (2015)
9. Chung, J., Hulbert, G.: A time integration algorithm for
structural dynamics with improved numerical dissipation:
the generalized-α method. J. Appl. Mech. 60, 371–375
(1993)
10. Dahlquist, G.G.: A special stability problem for linear mul-
tistep methods. BIT Numer. Math. 3(1), 27–43 (1963)
11. Dong, S.: BDF-like methods for nonlinear dynamic analysis.
J. Comput. Phys. 229(8), 3019–3045 (2010)
12. Fung, T.: Weighting parameters for unconditionally stable
higher-order accurate time step integration algorithms. part
1—first-order equations. Int. J. Numer. Methods Eng. 45(8),
941–970 (1999)
13. Fung, T.: Weighting parameters for unconditionally sta-
ble higher-order accurate time step integration algorithms.
Part 2–second-order equations. Int. J. Numer. Methods Eng.
45(8), 971–1006 (1999)
14. Fung, T.: Solving initial value problems by differential
quadrature method. part 1: first-order equations. Int. J.
Numer. Methods Eng. 50(6), 1411–1427 (2001)
15. Fung, T.: Solving initial value problems by differential
quadrature method. part 2: second-and higher-order equa-
tions. Int. J. Numer. Methods Eng. 50(6), 1429–1454 (2001)
16. Gear, C.W.: Numerical Initial Value Problems in Ordi-
nary Differential Equations. Prentice Hall PTR, New Jersey
(1971)
17. Hilber, H.M., Hughes, T.J., Taylor, R.L.: Improved numer-
ical dissipation for time integration algorithms in structural
dynamics. Earthq. Eng. Struct. Dyn. 5(3), 283–292 (1977)
18. Ji, Y., Xing, Y.: An optimized three-sub-step composite time
integration method with controllable numerical dissipation.
Comput. Struct. 231, 106210 (2020)
19. Kennedy, C.A., Carpenter, M.H.: Diagonally implicit
Runge–Kutta methods for stiff ODEs. Appl. Numer. Math.
146, 221–244 (2019)
20. Kim, W., Choi, S.Y.: An improved implicit time integration
algorithm: the generalized composite time integration algo-
rithm. Comput. Struct. 196, 341–354 (2018)
21. Kim, W., Reddy, J.: An improved time integration algorithm:
a collocation time finite element approach. Int. J. Struct.
Stab. Dyn. 17(02), 1750024 (2017)
22. Li, J., Yu, K., He, H.: A second-order accurate three sub-step
composite algorithm for structural dynamics. Appl. Math.
Model. 77, 1391–1412 (2020)
23. Li, J., Yu, K., Li, X.: A novel family of controllably dissipa-
tive composite integration algorithms for structural dynamic
analysis. Nonlinear Dyn. 96(4), 2475–2507 (2019)
24. Masarati, P., Lanz, M., Mantegazza, P.: Multistep integra-
tion of ordinary, stiff and differential-algebraic problems
for multibody dinamics applications. In: Xvi Congresso
Nazionale AIDAA, pp. 1–10 (2001)
25. Newmark, N.M.: A method of computation for structural
dynamics. J. Eng. Mech. Div. 85(3), 67–94 (1959)
26. Noh, G., Bathe, K.J.: The Bathe time integration method
with controllable spectral radius: the ρ∞ -Bathe method.
Comput. Struct. 212, 299–310 (2019)
123
1962
27. Noh, G., Ham, S., Bathe, K.J.: Performance of an implicit
time integration scheme in the analysis of wave propaga-
tions. Comput. Struct. 123, 93–105 (2013)
28. Tamma, K.K., Har, J., Zhou, X., Shimada, M., Hoitink,
A.: An overview and recent advances in vector and scalar
formalisms: space/time discretizations in computational
dynamics–a unified approach. Arch. Comput. Methods Eng.
18(2), 119–283 (2011)
29. Wood, W., Bossak, M., Zienkiewicz, O.: An alpha modifi-
cation of Newmark’s method. Int. J. Numer. Methods Eng.
15(10), 1562–1566 (1980)
30. Xie, Y., Steven, G.P.: Instability, chaos, and growth and
decay of energy of time-stepping schemes for non-linear
dynamic equations. Commun. Numer. Methods Eng. 10(5),
393–401 (1994)
31. Xing, Y., Zhang, H., Wang, Z.: Highly precise time integra-
tion method for linear structural dynamic analysis. Int. J.
Numer. Methods Eng. 116(8), 505–529 (2018)
123
H. Zhang et al.
32. Zhang, H., Xing, Y.: Optimization of a class of composite
method for structural dynamics. Comput. Struct. 202, 60–73
(2018)
33. Zhang, J.: A-stable two-step time integration methods with
controllable numerical dissipation for structural dynamics.
Int. J. Numer. Methods Eng. 121, 54–92 (2020)
34. Zhou, X., Tamma, K.K.: Design, analysis, and synthesis of
generalized single step single solve and optimal algorithms
for structural dynamics. Int. J. Numer. Methods Eng. 59(5),
597–668 (2004)
Publisher’s Note Springer Nature remains neutral with regard
to jurisdictional claims in published maps and institutional affil-
iations.